Brain functional connectivity correlation value clustering device, brain functional connectivity correlation value clustering system, brain functional connectivity correlation value clustering method, brain functional connectivity correlation value classifier program, brain activity marker classification system and clustering classifier model for brain functional connectivity correlation values

Abstract

A brain functional connectivity correlation value clustering device for clustering subjects having a prescribed attribute on the basis of brain measurement data obtained from a plurality of facilities, wherein a plurality of MRI devices capture resting state fMRI image data of a healthy cohort and a patient cohort; a computing system 300 performs generation of an identifier as ensemble learning of “supervised learning” between harmonized component values of correlation matrixes and disease labels of each of the subjects, selects, during the ensemble learning, features for clustering in accordance with importance from the features specified for generating an identifier for a disease label, and performs multiple co-clustering by “unsupervised learning.”

The invention claimed is: 
     
       1. A clustering device for clustering brain functional connectivity correlation values performing, based on measurement results of brain activities of objects, clustering of objects having at least one prescribed attribute among said objects, comprising
 a computing system for performing a process of said clustering based on measured values of brain activities, on a plurality of subjects including a first cohort of subjects having said prescribed attribute and a second cohort of subjects not having said prescribed attribute, said computing system including a storage device and a processor; wherein 
 said processor is configured to perform the steps of 
 i) storing, in said storage device, for each of said plurality of subjects, features based on a plurality of brain functional connectivity correlation values respectively representing time-wise correlation of brain activities between each of a prescribed plurality of pairs of brain regions, and 
 ii) based on said features stored in said storage device, conducting supervised machine learning for generating an identifier model for discriminating presence/absence of said attribute; 
 said processor performs, in the machine learning for generating an identifier model, the steps of 
 generating a plurality of training sub-samples by performing under-sampling and sub-sampling from said first cohort of subjects and said second cohort of subjects, and 
 selecting, for each of said training sub-samples, from a sum set of features used for generating the identifier by machine learning, features for clustering in accordance with importance of features belonging to said sum set; and 
 said processor further performs the step of clustering said first cohort of subjects through multiple co-clustering of unsupervised learning, based on the selected features for clustering, and thereby generates a cluster classifier. 
 
     
     
       2. The clustering device for clustering brain functional connectivity correlation values according to  claim 1 , receiving, from a plurality of brain activity measuring devices provided respectively at a plurality of measurement sites, information representing time-wise correlation of brain activities between each of a prescribed plurality of pairs of brain regions of each of a plurality of subjects; wherein
 said computing system includes 
 a harmonization calculating means for correcting said plurality of brain functional connectivity correlation values of each of said plurality of subjects to remove measurement bias of said measurement sites and thereby storing corrected adjusted values as said features in said storage device. 
 
     
     
       3. The clustering device for clustering brain functional connectivity correlation values according to  claim 1 , wherein
 the process of generating an identifier by the machine learning involves ensemble learning of generating a plurality of identifier sub-models respectively for said plurality of training samples, and integrating said plurality of identifier sub-models to generate said identifier model. 
 
     
     
       4. The clustering device for clustering brain functional connectivity correlation values according to  claim 1 , wherein
 said attribute is represented by a label of diagnosis result as having a prescribed psychiatric disorder; and 
 said clustering is a process of classifying, by data-driven machine learning, said first cohort of subjects into clusters of at least one subtype. 
 
     
     
       5. The clustering device for clustering brain functional connectivity correlation values according to  claim 1 , wherein
 in generating an identifier by said machine learning, said processor performs the steps of 
 i) dividing said adjusted values into a training dataset for machine learning and a test dataset for validation; 
 ii) performing under-sampling and sub-sampling a prescribed number of times on said training dataset to generate said prescribed number of training sub-samples; 
 iii) generating an identifier sub-model for each of said training sub-samples; and 
 iv) integrating outputs of said identifier sub-models and generating an identifier model regarding presence/absence of said attribute. 
 
     
     
       6. The clustering device for clustering brain functional connectivity correlation values according to  claim 1 , wherein
 the process of generating an identifier by said machine learning is nested cross-validation having external cross-validation and internal cross-validation; 
 in said nested cross-validation, said processor performs the steps of: 
 i) dividing said adjusted values into a training dataset for machine learning and a test dataset for validation by conducting K-fold cross-validation as said external cross-validation; 
 ii) performing under-sampling and sub-sampling a prescribed number of times on said training dataset to generate said prescribed number of training sub-samples; 
 iii) in each loop of said K-fold cross-validation, adjusting hyper-parameters by said internal cross-validation and thereby generating an identifier sub-model for each of said training sub-samples; and 
 iv) generating an identifier model regarding presence/absence of said attribute based on said identifier sub-models. 
 
     
     
       7. The clustering device for clustering brain functional connectivity correlation values according to  claim 3 , wherein
 the process of generating an identifier by said machine learning is machine learning with feature selection; and 
 in selecting a feature for the clustering, importance of a feature belonging to said sum set is determined by a ranking of frequency of said feature being selected when said identifier sub-model is selected. 
 
     
     
       8. The clustering device for clustering brain functional connectivity correlation values according to  claim 3 , wherein
 the process of generating an identifier by said machine learning is a random forest method; and 
 in selecting a feature for the clustering, importance of a feature belonging to said sum set is an importance calculated in accordance with Gini impurity in the random forest method for each feature. 
 
     
     
       9. The clustering device for clustering brain functional connectivity correlation values according to  claim 3 , wherein
 the process of generating an identifier by said machine learning is machine learning by L2 regularization; and 
 in selecting a feature for the clustering, importance of a feature belonging to said sum set is determined by a ranking based on feature weight in said identifier sub-model calculated by L2 regularization. 
 
     
     
       10. The clustering device for clustering brain functional connectivity correlation values according to  claim 2 , wherein:
 said storage device stores in advance, for a plurality of traveling subjects as common objects of measurements across said plurality of measurement sites, results of measurements of brain activities of a predetermined plurality of brain regions of each of said traveling subjects; 
 said processor performs the steps of 
 for each of said traveling subjects, calculating a prescribed component of a brain functional connectivity matrix representing time-wise correlation of brain activities of said plurality of pairs of brain regions; and 
 by using Generalized Linear Mixed Model, for each prescribed component of said functional connectivity matrix, calculating said measurement bias as a fixed effect at each measurement site with respect to an average of the component over said plurality of measurement sites and said plurality of traveling subjects. 
 
     
     
       11. The clustering device for clustering brain functional connectivity correlation values according to  claim 4 , wherein said processor performs said process of classifying into said subtypes based on measurement data of a subject measured at a measurement site other than said plurality of measurement sites. 
     
     
       12. A clustering system for clustering brain functional connectivity correlation values performing, based on measurement results of brain activities of objects, clustering of objects having at least one prescribed attribute among said objects, comprising
 a plurality of brain activity measuring devices provided respectively at a plurality of measurement sites, for time-sequentially measuring brain activities of a plurality of subjects including a first cohort of subjects having said prescribed attribute and a second cohort of subjects not having said prescribed attribute, and 
 a computing system for performing a process of said clustering based on measured values of brain activities, on said plurality of subjects, said computing system including a storage device and a processor; wherein 
 said processor is configured to perform the steps of 
 i) storing, in said storage device, for each of said plurality of subjects, features based on a plurality of brain functional connectivity correlation values respectively representing time-wise correlation of brain activities between each of a prescribed plurality of pairs of brain regions, and 
 ii) based on said features stored in said storage device, conducting supervised machine learning for generating an identifier model for discriminating presence/absence of said attribute; 
 said processor performs, in the machine learning for generating an identifier model, the steps of 
 generating a plurality of training sub-samples by performing under-sampling and sub-sampling from said first cohort of subjects and said second cohort of subjects, and 
 selecting, for each of said training sub-samples, from a sum set of features used for generating the identifier by machine learning, features for clustering in accordance with importance of features belonging to said sum set; and 
 said processor further performs the step of clustering said first cohort of subjects through multiple co-clustering of unsupervised learning, based on the selected features for clustering, and thereby generating a cluster classifier. 
 
     
     
       13. The clustering system for clustering brain functional connectivity correlation values according to  claim 12 , wherein said computing system receives, from a plurality of brain activity measuring devices provided respectively at a plurality of measurement sites, information representing time-wise correlation of brain activities between each of a prescribed plurality of pairs of brain regions of each of said plurality of subjects; and includes
 a harmonization calculating means for correcting said plurality of brain functional connectivity correlation values of each of said plurality of subjects to remove measurement bias of said measurement sites and thereby storing corrected adjusted values as said features in said storage device. 
 
     
     
       14. The clustering system for clustering brain functional connectivity correlation values according to  claim 12 , wherein
 said attribute is represented by a label of diagnosis result as having a prescribed psychiatric disorder; and 
 said clustering is a process of classifying, by data-driven machine learning, said first cohort of subjects into clusters of at least one subtype. 
 
     
     
       15. A clustering method of clustering brain functional connectivity correlation values allowing a computing system to perform, based on measurement results of brain activities of objects, clustering of objects having at least one prescribed attribute among said objects, wherein
 said computing system includes a storage device and a processor, 
 said method comprising the steps of: 
 said processor storing, in said storage device, for each of said plurality of subjects including a first cohort of subjects having said prescribed attribute and a second cohort of subjects not having said prescribed attribute, features based on brain functional connectivity correlation values representing time-wise correlation of brain activities between each of a prescribed plurality of pairs of brain regions, and 
 based on said features stored in said storage device, said processor conducting supervised machine learning for generating an identifier model for discriminating presence/absence of said attribute; wherein 
 said step of machine learning for generating an identifier model includes the steps of: 
 generating a plurality of training sub-samples by performing under-sampling and sub-sampling from said first cohort of subjects and said second cohort of subjects, and 
 selecting, for each of said training sub-samples, from a sum set of features used for generating the identifier by machine learning, features for clustering in accordance with importance of features belonging to said sum set; 
 said method further comprising the step of: 
 said processor clustering said first cohort of subjects through multiple co-clustering of unsupervised learning, based on the selected features for clustering, and thereby generating a cluster classifier. 
 
     
     
       16. A brain function marker classifying system, generated by a computing system performing, based on measurement results of brain activities of objects, clustering of objects having at least one prescribed attribute among said objects, allowing a computer to perform classifying of input data into clusters corresponding to the result of clustering, wherein
 said brain activity marker classifying system has a classifying function allowing the computer to classify into a cluster in which said input data has maximum posterior probability, based on a probability distribution model of each of said clusters; 
 said computing system includes a storage device and a processor; and 
 in the process of generating said brain activity marker classifying system based on said clustering, said computing system performs the steps of: 
 said processor storing, in said storage device, for each of a plurality of subjects including a first cohort of subjects having said prescribed attribute and a second cohort of subjects not having said prescribed attribute, features based on a plurality of brain functional connectivity correlation values representing time-wise correlation of brain activities between each of a prescribed plurality of pairs of brain regions, and 
 based on said features stored in said storage device, said processor conducting supervised machine learning for generating an identifier model for discriminating presence/absence of said attribute; wherein 
 said step of machine learning for generating an identifier model includes the steps of: 
 generating a plurality of training sub-samples by performing under-sampling and sub-sampling from said first cohort of subjects and said second cohort of subjects, and 
 selecting, for each of said training sub-samples, from a sum set of features used for generating the identifier by machine learning, features for clustering in accordance with importance of features belonging to said sum set; and wherein 
 said processor performs clustering of said first cohort of subjects through multiple co-clustering of unsupervised learning, based on the selected features for clustering, and thereby generates a cluster classifier. 
 
     
     
       17. The brain function marker classifying system according to  claim 16 , wherein said computing system performs the steps of:
 receiving, from a plurality of brain activity measuring devices provided respectively at a plurality of measurement sites, information representing time-wise correlation of brain activities between each of a prescribed plurality of pairs of brain regions of each of said plurality of subjects; and 
 conducting harmonization for correcting a plurality of brain functional connectivity correlation values representing time-wise correlation of said brain activities of each of said plurality of subjects to remove measurement bias of said measurement sites and thereby storing corrected adjusted values as said features in said storage device. 
 
     
     
       18. The brain function marker classifying system according to  claim 16 , wherein
 said attribute is represented by a label of diagnosis result as having a prescribed psychiatric disorder; and 
 said clustering is a process of classifying, by data-driven machine learning, said first cohort of subjects into clusters of at least one subtype. 
 
     
     
       19. A clustering classifier model for clustering brain functional connectivity correlation values generated by a computing system performing, based on measurement results of brain activities of objects, clustering of objects having at least one prescribed attribute among said objects, allowing a computer to perform classifying of input data into clusters corresponding to the result of clustering, wherein
 for each of views obtained by partitioning a group of features characterizing said object included in training data, said clustering classifier model has a function of classifying said input data into a cluster in which said input data has maximum posterior probability, in accordance with a value of probability density function calculated for said input data, based on information of said features included in each of said views and based on information specifying cluster-by-cluster probability density function of said object in each of said views; 
 said computing system includes a storage device and a processor; and 
 in the process of generating said clustering classifier model based on said clustering, said computing system performs the steps of: 
 said processor storing, in said storage device, for each of a plurality of subjects including a first cohort of subjects having said prescribed attribute and a second cohort of subjects not having said prescribed attribute, features based on a plurality of brain functional connectivity correlation values representing time-wise correlation of brain activities between each of a prescribed plurality of pairs of brain regions, and 
 based on said features stored in said storage device, said processor conducting supervised machine learning for generating an identifier model for discriminating presence/absence of said attribute; wherein 
 said step of machine learning for generating an identifier model includes the steps of: 
 generating a plurality of training sub-samples by performing under-sampling and sub-sampling from said first cohort of subjects and said second cohort of subjects, and 
 selecting, for each of said training sub-samples, from a sum set of features used for generating the identifier by machine learning, features for clustering in accordance with importance of features belonging to said sum set; and wherein 
 said processor performs clustering of said first cohort of subjects through multiple co-clustering of unsupervised learning, based on the selected features for clustering, partitions said features to said views, and generates the cluster-by-cluster probability density function of said object in each of said views.